Conventionally, there is known a lane mark recognizer which recognizes lane marks for dividing a traffic lane provided on a road on the basis of image data acquired by an in-vehicle camera which captures an image of the road ahead of the vehicle. As the lane marks provided on the road, there are used linear lane marks such as lane boundary lines (white lines) and stud-type lane marks which are discretely provided such as Botts Dots or cat's eyes.
Note that the linear lane marks can be detected by a straight line extraction technique such as a Hough transform while the stud-type lane marks cannot be detected by the straight line extraction technique. Therefore, there has been suggested a lane mark recognizer capable of detecting stud-type lane marks by using a pattern matching technique (for example, refer to Japanese Laid-Open No. 2003-317106).
The above lane mark recognizer, however, has a problem in that in some cases it cannot detect stud-type lane marks due to an increase in a difference between the shape of the stud-type lane marks in image data and a reference shape for pattern matching when the road image data is captured in cases where a part of the stud-type lane marks is broken or depending on how the light impinges on the stud-type lane marks.